North America 20k

The Americas through deep time

Ice sheets, sea level, vegetation from pollen, present-day plant ranges, floods and lakes, archaeological sites, languages, plant use and European arrivals from Greenland to Tierra del Fuego: a 35,000-year slider, the last 2,000 years in detail, and ice-sheet stages back to 3.6 million years. Ages are calendar years before 1950 (ka = thousand years, Ma = million).

20.0 ka18,050 BCE

The Americas, Greenland to Tierra del Fuego

Two fingers to move or zoom the map; tap a point for details

Salish Sea and Orcas Island

rebound, subduction, AD 1700

Why the shoreline here moves differently from the global curve. The Cordilleran ice sheet pressed the crust down by a few hundred metres. When the Puget and Juan de Fuca lobes melted back (about 16 ka), the sea flooded the depressed land, leaving marine deposits and shorelines high on the San Juan Islands. The crust then rebounded fast, within a few thousand years, and local sea level fell below today's before the global rise caught up. The blue line is the global (ice volume) curve; the orange line is a schematic local curve for the San Juan Islands with illustrative values (shape after Clague & James 2002, James et al. 2009, Dethier et al. 2016). Land and water on the map are drawn from ETOPO1 topography and that curve, so the drowned area is only as good as the curve.

Cascadia, 26 January 1700. The Juan de Fuca plate subducts beneath North America at about 3–4 cm/yr. The last great rupture (about M9) dropped the outer coast by 0.5–2 m in minutes, so tides killed coastal spruce and cedar forests that still stand as "ghost forests" (Copalis River, Niawiakum River). Tree rings date the deaths to the 1699 growing season or soon after (Yamaguchi et al. 1997). Japanese records of a tsunami with no felt earthquake on 27–28 January 1700 fix the date (Satake et al. 1996; Atwater et al. 2005). Move the slider to 0.25 ka to see it.

Who moved first: life histories from the silvics

click a row to map that taxon

Traits summarised from Silvics of North America (Burns & Honkala 1990, USDA Agriculture Handbook 654) for the eastern species most likely behind each pollen type. Values are typical, rounded, and for the named species only. The strategy column is an interpretive grouping: sprint colonists (small wind-borne seeds, early reproduction, shade-intolerant), marathon species (shade-tolerant, slow to mature, long-lived, slow advance), and bet-hedgers (spread risk through masting, serotiny, sprouting, layering or seed dormancy). The chart plots each taxon's abundance-weighted mean latitude east of 100°W from the Williams et al. (2004) grids, so you can compare the grouping with how fast each pollen type actually moved.

Anomalies worth checking against the site records. At 17–18 ka a band of cells at 41–46°N (Iowa–Nebraska, by the ice margin) carries about 1.6% Acer, each interpolated from only 2–4 sites, while sites further south sit just under 1%. Site records explain it: the band traces back to Wolf Creek, Minnesota, where the maple is box elder (A. negundo, 1–2.4% at about 20–17 ka) alongside abundant ragweed and sage, either floodplain pioneers or older pollen reworked into the lake. Meanwhile Cupola Pond in the Ozarks, with a tight chronology, has sugar maple at a steady 0.3–0.6% (Neotoma datasets 3029, 47555, 47557). Cupressaceae/Taxaceae at ≥5% in the East is almost all on the Virginia–Carolina coastal plain (most likely Chamaecyparis thyoides and Juniperus virginiana), and the grids' own Thuja column is nearly empty. North of 42°N there is a real but small signal: cells at ≥2% (about 2.5–3%) are centred in the Maritimes at 12 ka and in northern Ontario, north of Lake Superior, by 10–8 ka, following the retreating ice. That fits Thuja on fresh calcareous till or juniper on open ground; the pollen type cannot separate them.

Plant-use communities across language lines

Native American Ethnobotany database
Loading…

Does language ancestry predict plant use?

Each of the peoples in the plant-use analysis is placed on the Glottolog 5.3 language tree, used here as an a priori hypothesis: if plant knowledge is inherited with language, peoples whose languages split recently should use similar plants, wherever they live. Tree distance is how deep in the tree two languages part (Glottolog gives topology, not dates, so depth stands in for time since divergence).

Plant-use dissimilarity between every pair of peoples, both axes ordered by the language tree (families by size, then branch by branch). Dark = similar. Dark blocks on the diagonal mean related peoples use similar plants; dark patches off the diagonal mean unrelated peoples do.

How sure is this? A ledger of doubt

Every layer mixes measurement, interpolation and drawing by hand. This ledger sorts the main weak points into five kinds, from quantified error to popular accounts that mislead if read literally. The same colours mark the "How sure" lines in At this time and the pins in the doubt overlay.

measured errorKnown, quantified

Ice margins. NADI-1 gives optimal, minimum and maximum margins. Where few dates exist (interior Keewatin, Hudson Bay, the Cordilleran interior) the range spans tens to hundreds of kilometres at the same time slice. The doubt overlay hatches that band.Dalton et al. 2023
Ages. Dated samples carry ±1–2 centuries; interpolated full-glacial sample ages are much looser. At Hansen Marsh, Wisconsin, a sample modelled at 19.5 ka has a range of 16.1–23.9 ka.Neotoma dataset 1004, PalEON-STEPPS chronology
Sea level. LGM global sea level estimates run from about −120 to −135 m; the size of meltwater pulse 1A from about 8 to 20 m.Lambeck et al. 2014 and studies reviewed there
Relief. Topography is sampled at 1/3° (about 37 km), so narrow straits, small islands and shelf edges are blurred.ETOPO1

interpolatedFilled in between data

Pollen grids. About 500 cores interpolated to a 50 km grid. At 18–20 ka the median cell rests on 3 sites (11 today); cells from fewer than 3 sites are pale on the map and hatched in the doubt overlay.Williams et al. 2004
One site can paint a region. The 17–19 ka Acer band in Iowa–Nebraska traces back to Wolf Creek, Minnesota, where the maple is box elder (A. negundo, 1–2.4%) with abundant ragweed and sage. The grid spread it south-west where no other sites exist.Neotoma dataset 3029 (Birks 1976)
Time blending. Ice is blended between 0.5 ka isochrones and pollen between 1 ka slices, so events shorter than about 500 years (the 8.2 ka event, the Younger Dryas onset) are smoothed out.this page
Pollen is not trees. Pine and birch are over-represented; poplar, larch and maple under-represented. Types are lumped: Acer merges sugar maple with box elder; Cupressaceae merges Thuja, juniper and Atlantic white-cedar.Williams et al. 2004; Neotoma taxonomy
Language families. Each Glottolog language has one point, often where it was documented or is spoken today. The shading assigns land to the nearest point within 300 km; real boundaries were fuzzy, multilingual and moved, and many peoples were displaced after 1600.Glottolog 5.3
Plant-use records. The ethnobotany database compiles about 200 published ethnographies, mostly of 1880–1990. Coverage is uneven (California, the Northwest Coast and the Southwest are heavily documented; the Southeast and Mexico thinly), names are those the sources used, and many peoples had been moved from their homelands before the records were made. Each people is placed by hand at an approximate homeland point.Moerman 1998; naeb.brit.org
Where the work has been done. The density veil shows how unevenly each field has looked: 2,121 Neotoma pollen sites, only 240 south of 15°N; 73,093 radiocarbon dates in the Americas, 650 of them older than 13,000 radiocarbon years; languages scored by how many survey catalogues list them. Blank areas on the other layers often mean nobody looked.Neotoma; Bird et al. 2022 (p3k14c); Glottolog 5.3
Latin American vegetation. No gridded pollen maps like Williams et al. exist south of Mexico; the map shows BIOME 6000 site biomes at 0, 6 and 21 ka instead, each from one site's pollen.Harrison 2017; Marchant et al. 2009
Spread isochrones. Each front is interpolated (inverse-distance weighting, within 700 km of a dated record, on land) from a handful of curated earliest dates, so the lines show the shape of the published story, not a measured surface. Earliest dates move as new sites are found, and several traits (pottery, possibly maize use) began in more than one place.sources per trait in the legend and tooltips
Ranges read from pictures. BONAP county status was read by colour from each map image. A few tiny counties and independent cities (Virginia cities, Washington DC, small New York boroughs) are too small to read and show the state's value. Questionable (cross-hatched) records read as whichever colour dominates.BONAP 2014; this page
Ranges are today's, samples span time. The county maps show where a plant grows now; a site that records it may date from 15,000 years ago, when the plant grew somewhere else. Names also differ: a pollen "type" can cover several species, and Moerman's names are matched to BONAP by spelling only.this page
Language loss. "Last speaker" dates mark the last known first-language speaker and are approximate; communities are reviving several of these languages, and many more Indigenous languages are endangered today.
Database chronologies can be wrong. Dark Lake and Little Pine Lake, Wisconsin, are varved records about 600–990 years long, yet a stored age model gives their basal samples ages of 15–41 ka. A search by age returns them as full-glacial sites.Neotoma datasets 546, 1643

schematicDrawn by hand

Lake Agassiz stages and Lake Missoula. Outlines and dates approximate, traced from published reconstructions.after Teller & Leverington 2004; Smith 2006
Flood paths and pulses, the 8.2 ka arrow, Cascadia trench and ridges. Routes are real; the drawn lines, timing and spacing are illustrative.
Deep-time ice sheets. Batchelor et al. (2019) give a best estimate with minimum and maximum versions for each stage; for the late Gauss Chron the area runs from 3.8 to 20.1 million km² around a best estimate of 11.0. Only the best estimates are drawn. Sea level before 798 ka has no stack and is not shown.
Salish Sea sea-level curve. Shape from published curves, values illustrative (±20 m or more).
Sea level without isostasy. The continental map floods today's topography; near former ice the real shoreline differed by tens to hundreds of metres.
Life-history grouping. Sprint / marathon / bet-hedger is an interpretation of silvics traits, not a category in the handbook.

debatedExperts disagree

What triggered the Younger Dryas. Agassiz water to the St Lawrence (Broecker et al. 1989), or to the Arctic via the Mackenzie (Murton et al. 2010; Keigwin et al. 2018), since the eastern outlet may still have been blocked (Lowell et al. 2005). The comet-impact idea (Firestone et al. 2007) is rejected by most specialists (Holliday et al. 2023).
How many Missoula floods. About 40 separate jökulhlaups (Waitt 1980, 1985) versus fewer, larger floods with subglacial sources (Shaw et al. 1999).
The ice-free corridor. Open by about 14 ka, but plants and animals may not have made it passable until about 12.6 ka (Pedersen et al. 2016; Heintzman et al. 2016). Whether the first people used it or a Pacific coast route is open; White Sands footprints dated 23–21 ka (Bennett et al. 2021; Pigati et al. 2023) place people south of the ice before either route opened, and that dating has itself been debated.
How fast trees moved. Pollen maps imply rapid spread. Genetic evidence puts small populations of red maple and beech close to the ice, which would mean slower spread from nearer refuges (McLachlan et al. 2005). The Wolf Creek box elder belongs to this question.
How many people lived here in 1492. Estimates for north of Mexico run from about 2 to 18 million; for the Americas, about 40 to 100 million. How much of the collapse came from disease, versus war, enslavement and displacement, is argued.
Cascadia recurrence. Turbidite records suggest 19+ full-margin ruptures in 10,000 years (Goldfinger et al. 2012); whether turbidites record full ruptures is questioned (Atwater et al. 2014).

myth checkPopular accounts, read literally

"North America was frozen." At the LGM ice covered Canada and the northern US; the Southeast held pine and oak woods, and Beringia was steppe-tundra.
"The ice was two miles thick." True only near the ice domes; thickness fell to a few hundred metres at the lobes and to zero at the margin.
"A narrow land bridge." Beringia was a lowland about 1,000 km from north to south, a region people and animals lived in, not a causeway.
"Forests marched north." Species moved separately, at different speeds and directions, forming mixes with no modern equivalent, such as spruce with ash and hop-hornbeam (Williams et al. 2004).
"Sea level was 120 m lower." That is the global ice-volume number. Locally, rebound and subsidence shifted shorelines by tens to hundreds of metres (see the Salish Sea panel).
"The Great Lakes were dug by the ice." Ice deepened basins that followed older river valleys and weak bedrock; it did not create them from nothing.
"Cascadia is overdue." An average interval does not make a rupture due. Intervals vary from a few centuries to about 1,000 years; the hazard is real at any time.
"An untouched wilderness." Indigenous peoples burned, farmed, built cities (Cahokia, Teotihuacan, Tenochtitlan) and managed forests and prairies for millennia; much "pristine" land in 1800s accounts had been emptied by epidemics.
"Columbus discovered America." Millions of people already lived here, and Norse Greenlanders had reached Newfoundland by 1021.
"Clovis first." Long taught as fact; several well-dated sites are older than Clovis, though how much older is still argued.

Sources, and what is drawn by hand

Ice margins (data). Dalton, A.S. et al. 2023. Deglaciation of the North American Ice Sheet Complex in calendar years based on a comprehensive database of chronological data: NADI-1. Quaternary Science Reviews 321, 108345. Shapefiles: doi.org/10.5281/zenodo.8161764 (CC-BY-4.0), downloaded 8 Nov 2026. NADI-1 updates Dalton et al. 2020 (QSR 234, 106223), which updated Dyke 2004. Optimal margins drawn every 0.5 ka and cross-faded; dashed and dotted outlines are the NADI-1 maximum and minimum margins. Polygons simplified to about 5 km. NADI-1 ends at 1 ka; the last frame is held to the present.

Vegetation (data). Williams, J.W., Shuman, B.N., Webb, T. III, Bartlein, P.J. & Leduc, P.L. 2004. Late Quaternary vegetation dynamics in North America: scaling from taxa to biomes. Ecological Monographs 74, 309–334. Gridded pollen percentages (about 50 km grid, 1 ka steps, interpolated from about 500 North American Pollen Database / Neotoma cores), NOAA/WDS Paleoclimatology data contribution 2003-089, downloaded 8 Nov 2026. Stored at reduced precision (round(1.5·√%)), so small percentages are binned. Linear blending between 1 ka slices.

Pollen limits to keep in mind. Cupressaceae/Taxaceae pollen stands for Thuja but also Juniperus and Taxus. Pinus banksiana and P. resinosa pollen cannot be told apart; only subgenus Strobus (white pine) is separable, and many records lump all pine. Larix is under-represented (low production, poor preservation) and is merged with Pseudotsuga here. Populus pollen is fragile and badly under-represented. Pine and birch are over-represented because they produce and disperse so much pollen. The 0 ka slice comes from core tops and can include post-settlement Ambrosia. Why pollen percentages are not forest cover: try the pollen-rain simulator →

Tension zone (data-derived). Line where southern hardwood pollen (oak, hickory, elm, ash, hop-hornbeam, chestnut, maple) equals northern conifer and birch pollen (pine, hemlock, spruce, fir, birch, Cupressaceae). Today it runs near 44°N in Lower Michigan and Wisconsin, close to the mapped floristic tension zone. It is a pollen proxy, not the floristic boundary itself.

Topography and sea level (data + curve). ETOPO1 1 arc-minute relief (Amante & Eakins 2009, NOAA NGDC) via NOAA CoastWatch ERDDAP (dataset etopo180), sampled at 1/3° for the continent and 1′ for the Salish Sea, downloaded 8 Nov 2026. Coastlines are today's topography flooded to a global mean sea level curve approximated from Lambeck et al. 2014 (PNAS 111, 15296). Isostatic depression and forebulges are ignored on the continental map, so formerly ice-covered coasts (Hudson Bay, Champlain Sea, Salish Sea) are wrong in detail. Present lakes and coastline: Natural Earth 1:50m (public domain).

Proglacial lakes (schematic). Lake Agassiz stages (Lockhart, Moorhead low, Emerson, Nipigon, Ojibway) are hand-drawn after Teller & Leverington 2004 and Leverington et al. 2002; extents and ages are approximate. The final drainage arrow (about 8.2 ka) follows Barber et al. 1999 and Lajeunesse & St-Onge 2008. Glacial Lake Missoula is hand-drawn after Pardee 1942 and Smith 2006.

Missoula floods (schematic). Flood paths follow the routes mapped by Bretz (1923), Baker (1973) and O'Connor & Baker (1992): Rathdrum Prairie, Cheney–Palouse, Telford–Crab Creek and Grand Coulee tracts, Pasco Basin, Wallula Gap, Columbia Gorge, Willamette backflooding. Pulses mark roughly 40 floods between about 18.2 and 15.5 ka (Waitt 1985; Benito & O'Connor 2003; Balbas et al. 2017); their timing and spacing are illustrative.

Cascadia (schematic geometry, dated event). Trench and ridge traces are hand-drawn. Event: Atwater, B.F. et al. 2005, The Orphan Tsunami of 1700 (USGS Professional Paper 1707); Satake et al. 1996 (Nature 379, 246); Yamaguchi et al. 1997 (Nature 389, 922). Ghost-forest sites are approximate locations.

Salish Sea relative sea level (schematic). Curve shape after Clague & James 2002 (QSR 21, 71), James et al. 2009 (QSR 28, 1200) and Dethier et al. 2016 (GSA Bulletin); values are illustrative (±20 m or more), not a fit to dated samples.

Wider map, borders and Greenland (data). Relief: ETOPO1 sampled at 1/6° (5–85°N, 160°E–10°W) via NOAA CoastWatch ERDDAP, resampled to 8 km. Country, state and province lines and the Greenland ice outline: Natural Earth 1:50m (public domain), downloaded 8 Nov 2026. Greenland ice is today's outline at all times; NADI-1 does not cover Greenland, and at the LGM its ice reached the shelf edge (not drawn).

Languages (data, interpolated shading). Hammarström, H., Forkel, R., Haspelmath, M. & Bank, S. Glottolog 5.3 (2026), doi.org/10.5281/zenodo.21681980, CC-BY-4.0: language points and families for languages spoken in North and Central America, the Caribbean, Greenland and Chukotka. Sign languages, pidgins and unclassified entries left out. Shown from about 1350 CE because the points reflect documentation of about 1500–1900 CE and later.

Archaeological sites, oral histories and European settlements (curated). About 40 widely published sites with dates from the cited studies (shown in tooltips), placed to about 0.1°; no unpublished or sensitive locations. Oral-history markers: Klamath accounts associated with Mount Mazama (7.7 ka); coastal accounts of the 1700 earthquake (Ludwin et al. 2005, Seismological Research Letters 76, 140). European settlements by founding year from standard histories; L'Anse aux Meadows after Kuitems et al. 2021 (Nature 601, 388).

Plant use (data, curated placement). Moerman, D.E. 1998. Native American Ethnobotany. Timber Press; online database hosted by the Botanical Research Institute of Texas (naeb.brit.org), tribe pages downloaded 8 Nov 2026. Only tribe, species, use category and subcategory were kept; no note text is reproduced. Places are hand-assigned approximate homeland points (about 0.5–1°); regional labels such as "California Indian" are shown hollow and left out of clustering and centroids. Communities: genus presence per people, βsim dissimilarity (Lennon et al. 2001), average-linkage clustering, indicator values after Dufrêne & Legendre (1997), Mantel tests with 999 permutations.

The Americas, deep time and data density (8–9 Nov 2026). Display projection: spherical Lambert azimuthal equal-area centred at 15°N 85°W, relief from ETOPO1 at 1/6° (57°S–85°N) resampled to 10 km. Patagonian Ice Sheet: Davies, B.J. et al. 2020, Earth-Science Reviews 204, 103152 (PATICE), Mendeley Data doi 10.17632/rjnw46vbdw (CC BY-NC 3.0), at 35, 30, 25, 20, 15, 13, 10 and 5 ka and the Little Ice Age, with glacial lakes and margin uncertainty. Deep-time and 26.5–35 ka Northern Hemisphere ice: Batchelor, C.L. et al. 2019, Nature Communications 10, 3713, best-estimate outlines as repackaged in Hotham 2026 (Zenodo 23878610, CC-BY-4.0). Sea level beyond 21 ka: Spratt, R.M. & Lisiecki, L.E. 2016, Climate of the Past 12, 1079 (long PC1 stack, joined to the Lambeck curve at 21 ka). Site biomes: Harrison, S.P. 2017, BIOME 6000 DB classified plotfile v1, University of Reading, doi 10.17864/1947.99 (Latin America after Marchant et al. 2009, Climate of the Past 5, 725). Data density: Neotoma pollen site index (api.neotomadb.org, 9 Nov 2026); Bird, D. et al. 2022, p3k14c, Scientific Data 9, 27 (R package data, MIT); Glottolog 5.3 archive pages (Zenodo 24035415), counting which of seven catalogues (ELCat, WALS, MultiTree, Moseley & Asher, Ruhlen, the Hammarström bibliography, ISO 639-3) list each language. Densities are Gaussian-smoothed (σ = 150 km) on a 50 km grid and log-scaled.

Present-day plant ranges (data, read from map images). Kartesz, J.T., The Biota of North America Program (BONAP). 2015. North American Plant Atlas (bonap.net/napa), county distribution maps generated 2014 (© BONAP), downloaded 13 Nov 2026. BONAP publishes the maps only as images, so each county's status was read from the colour at that county on the species or genus map (Census 2014 county outlines fitted to the map, mean misfit under 1 pixel), following the BONAP colour key. Lower 48 states by county; Canadian provinces by province (only the southern part of each appears on the maps); Alaska, Hawaii, Mexico and the Caribbean are not covered. For educational and research use; cite BONAP, not this page, for any range.

Fossil and modern sample sites (data). Neotoma Paleoecology Database (api.neotomadb.org, CC-BY 4.0), downloaded 13 Nov 2026: every pollen, pollen surface sample and plant macrofossil dataset in the Americas, with each collection unit's depositional environment (grouped as lake, peat and mire, biological, other) and its taxon list. Taxon names are Neotoma's; families from Neotoma's hierarchy and the GBIF backbone.

Species-level plant use (data). Moerman 1998 via naeb.brit.org, all tribe pages downloaded 13 Nov 2026: tribe, species, use category and subcategory only, no notes. Names are matched to BONAP by binomial; older synonyms may miss a map.

Region overlays. Ecoregions: Dinerstein, E. et al. 2017, An ecoregion-based approach to protecting half the terrestrial realm, BioScience 67, 534 (RESOLVE Ecoregions 2017, CC-BY 4.0), rasterised at 10 km. Nested language groups: Glottolog 5.3 classification (family, branch, sub-branch), shaded by nearest language point within 300 km, an interpolation. Culture areas (schematic): grown from hand-placed seed points after Kroeber, A.L. 1939, Cultural and Natural Areas of Native North America (University of California Press), the area scheme of the Handbook of North American Indians (Sturtevant, gen. ed., 1978–), Kirchhoff, P. 1943 (Mesoamerica) and Steward, J.H. (ed.) 1946–59, Handbook of South American Indians; Kroeber's grand areas are approximated and his 84 sub-areas are not reproduced. Agreement: adjusted Rand index (Hubert & Arabie 1985) and normalised mutual information between the NAEB plant-use communities and each scheme.

Spread isochrones and language loss (curated, interpolated). Dated earliest records compiled by hand from: maize — Piperno et al. 2009 (PNAS 106, 5019), Piperno & Flannery 2001, Pohl et al. 2007, Grobman et al. 2012, Zarrillo et al. 2008, Merrill et al. 2009, Riley et al. 1994, Crawford et al. 1997, Hart & Lovis 2013; pottery — Roosevelt et al. 1991, Oyuela-Caycedo 1995, Sassaman 1993, Clark & Gosser 1995; bow and arrow — Blitz 1988, Bettinger & Eerkens 1999; horses — Haines 1938 and Taylor et al. 2023 (Science 379, 1316); Thule — Friesen & Arnold 2008; Apachean — Seymour 2012; Numic — Bettinger & Baumhoff 1982; Caribbean — Rouse 1992, Keegan & Hofman 2017, Fitzpatrick 2015; Tupi-Guarani — Noelli 1998, Bonomo et al. 2015; Inca — Rowe 1946, D'Altroy 2014; Aztec — Berdan & Anawalt 1992. Dates are approximate and rounded; check the primary source before citing a number.

Ways of reckoning. Anno Mundi after the Seder Olam count (AM 1 = 3761 BCE); Ussher, Annales 1650 (creation 4004 BCE, Flood 2348 BCE); Hijri years converted at 354.37 days per lunar year; Maya Long Count with the GMT correlation (13.0.0.0.0 = 11 August 3114 BCE); Sumerian King List and Berossus; Herodotus 4.5–10 for the Scythian account. Conversions are approximate to about a year.

Life-history traits. Burns, R.M. & Honkala, B.H. (tech. coords.) 1990. Silvics of North America, Vols 1–2. USDA Forest Service Agriculture Handbook 654. Numbers are rounded from the species chapters and are typical values, not ranges; check the chapter before citing a number.

🚧 Under construction